]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
The basic idea behind cross memory attach is to allow MPI programs doing
authorChristopher Yeoh <cyeoh@au1.ibm.com>
Wed, 3 Aug 2011 00:52:48 +0000 (10:52 +1000)
committerStephen Rothwell <sfr@canb.auug.org.au>
Wed, 10 Aug 2011 01:50:38 +0000 (11:50 +1000)
intra-node communication to do a single copy of the message rather than a
double copy of the message via shared memory.

The following patch attempts to achieve this by allowing a destination
process, given an address and size from a source process, to copy memory
directly from the source process into its own address space via a system
call.  There is also a symmetrical ability to copy from the current
process's address space into a destination process's address space.

- Use of /proc/pid/mem has been considered, but there are issues with
  using it:
  - Does not allow for specifying iovecs for both src and dest, assuming
    preadv or pwritev was implemented either the area read from or
  written to would need to be contiguous.
  - Currently mem_read allows only processes who are currently
  ptrace'ing the target and are still able to ptrace the target to read
  from the target. This check could possibly be moved to the open call,
  but its not clear exactly what race this restriction is stopping
  (reason  appears to have been lost)
  - Having to send the fd of /proc/self/mem via SCM_RIGHTS on unix
  domain socket is a bit ugly from a userspace point of view,
  especially when you may have hundreds if not (eventually) thousands
  of processes  that all need to do this with each other
  - Doesn't allow for some future use of the interface we would like to
  consider adding in the future (see below)
  - Interestingly reading from /proc/pid/mem currently actually
  involves two copies! (But this could be fixed pretty easily)

As mentioned previously use of vmsplice instead was considered, but has
problems.  Since you need the reader and writer working co-operatively if
the pipe is not drained then you block.  Which requires some wrapping to
do non blocking on the send side or polling on the receive.  In all to all
communication it requires ordering otherwise you can deadlock.  And in the
example of many MPI tasks writing to one MPI task vmsplice serialises the
copying.

There are some cases of MPI collectives where even a single copy interface
does not get us the performance gain we could.  For example in an
MPI_Reduce rather than copy the data from the source we would like to
instead use it directly in a mathops (say the reduce is doing a sum) as
this would save us doing a copy.  We don't need to keep a copy of the data
from the source.  I haven't implemented this, but I think this interface
could in the future do all this through the use of the flags - eg could
specify the math operation and type and the kernel rather than just
copying the data would apply the specified operation between the source
and destination and store it in the destination.

Although we don't have a "second user" of the interface (though I've had
some nibbles from people who may be interested in using it for intra
process messaging which is not MPI).  This interface is something which
hardware vendors are already doing for their custom drivers to implement
fast local communication.  And so in addition to this being useful for
OpenMPI it would mean the driver maintainers don't have to fix things up
when the mm changes.

There was some discussion about how much faster a true zero copy would
go. Here's a link back to the email with some testing I did on that:

http://marc.info/?l=linux-mm&m=130105930902915&w=2

There is a basic man page for the proposed interface here:

http://ozlabs.org/~cyeoh/cma/process_vm_readv.txt

This has been implemented for x86 and powerpc, other architecture should
mainly (I think) just need to add syscall numbers for the process_vm_readv
and process_vm_writev. There are 32 bit compatibility versions for
64-bit kernels.

For arch maintainers there are some simple tests to be able to quickly
verify that the syscalls are working correctly here:

http://ozlabs.org/~cyeoh/cma/cma-test-20110718.tgz

Signed-off-by: Chris Yeoh <yeohc@au1.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: James Morris <jmorris@namei.org>
Cc: <linux-man@vger.kernel.org>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
14 files changed:
arch/powerpc/include/asm/systbl.h
arch/powerpc/include/asm/unistd.h
arch/x86/include/asm/unistd_32.h
arch/x86/kernel/syscall_table_32.S
fs/aio.c
fs/compat.c
fs/read_write.c
include/linux/compat.h
include/linux/fs.h
include/linux/syscalls.h
mm/Makefile
mm/process_vm_access.c [new file with mode: 0644]
security/keys/compat.c
security/keys/keyctl.c

index f6736b7da463508e2b9777e851d2bdd2c597fec8..525db4b590c783ca171ad941d705907cd74796d8 100644 (file)
@@ -354,3 +354,5 @@ COMPAT_SYS_SPU(clock_adjtime)
 SYSCALL_SPU(syncfs)
 COMPAT_SYS_SPU(sendmmsg)
 SYSCALL_SPU(setns)
+COMPAT_SYS(process_vm_readv)
+COMPAT_SYS(process_vm_writev)
index b8b3f599362b70167ab8db7c49038ad08f2b02f2..d3d1b5efd7eb1204405fde26c701e7c52b925fd9 100644 (file)
 #define __NR_syncfs            348
 #define __NR_sendmmsg          349
 #define __NR_setns             350
+#define __NR_process_vm_readv  351
+#define __NR_process_vm_writev 352
 
 #ifdef __KERNEL__
 
-#define __NR_syscalls          351
+#define __NR_syscalls          353
 
 #define __NR__exit __NR_exit
 #define NR_syscalls    __NR_syscalls
index 593485b38ab38cc501e3f975209aabfb3f710106..599c77d38f332f06fdf1e34f7449e2b7b5c81203 100644 (file)
 #define __NR_syncfs             344
 #define __NR_sendmmsg          345
 #define __NR_setns             346
+#define __NR_process_vm_readv  347
+#define __NR_process_vm_writev 348
 
 #ifdef __KERNEL__
 
-#define NR_syscalls 347
+#define NR_syscalls 349
 
 #define __ARCH_WANT_IPC_PARSE_VERSION
 #define __ARCH_WANT_OLD_READDIR
index fbb0a045a1a23bc9bdc2f9dd23c6c9673e2e13f7..ca0fe4d1fe28a0c086979dcff83c4a3f4b9cf845 100644 (file)
@@ -346,3 +346,5 @@ ENTRY(sys_call_table)
        .long sys_syncfs
        .long sys_sendmmsg              /* 345 */
        .long sys_setns
+       .long sys_process_vm_readv
+       .long sys_process_vm_writev
index e29ec485af255822b8414be128fc8ef66da9a204..632b235f4fbe02c237de17fbde271c8bfe6f1240 100644 (file)
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -1387,13 +1387,13 @@ static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat)
                ret = compat_rw_copy_check_uvector(type,
                                (struct compat_iovec __user *)kiocb->ki_buf,
                                kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
-                               &kiocb->ki_iovec);
+                               &kiocb->ki_iovec, 1);
        else
 #endif
                ret = rw_copy_check_uvector(type,
                                (struct iovec __user *)kiocb->ki_buf,
                                kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
-                               &kiocb->ki_iovec);
+                               &kiocb->ki_iovec, 1);
        if (ret < 0)
                goto out;
 
index 0b48d018e38adbc1a72f78a4ab851626f2fffca8..53544759e42f766ccdf4380e509e437e7293c4f0 100644 (file)
@@ -550,7 +550,7 @@ out:
 ssize_t compat_rw_copy_check_uvector(int type,
                const struct compat_iovec __user *uvector, unsigned long nr_segs,
                unsigned long fast_segs, struct iovec *fast_pointer,
-               struct iovec **ret_pointer)
+               struct iovec **ret_pointer, int check_access)
 {
        compat_ssize_t tot_len;
        struct iovec *iov = *ret_pointer = fast_pointer;
@@ -597,7 +597,8 @@ ssize_t compat_rw_copy_check_uvector(int type,
                }
                if (len < 0)    /* size_t not fitting in compat_ssize_t .. */
                        goto out;
-               if (!access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
+               if (check_access &&
+                   !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
                        ret = -EFAULT;
                        goto out;
                }
@@ -1111,7 +1112,7 @@ static ssize_t compat_do_readv_writev(int type, struct file *file,
                goto out;
 
        tot_len = compat_rw_copy_check_uvector(type, uvector, nr_segs,
-                                              UIO_FASTIOV, iovstack, &iov);
+                                              UIO_FASTIOV, iovstack, &iov, 1);
        if (tot_len == 0) {
                ret = 0;
                goto out;
index 179f1c33ea578481cdc3388d54717089704a51a3..7a220bec217cf88b63947c63c4ac866b91b0ba45 100644 (file)
@@ -617,7 +617,8 @@ ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
                              unsigned long nr_segs, unsigned long fast_segs,
                              struct iovec *fast_pointer,
-                             struct iovec **ret_pointer)
+                             struct iovec **ret_pointer,
+                             int check_access)
 {
        unsigned long seg;
        ssize_t ret;
@@ -673,7 +674,8 @@ ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
                        ret = -EINVAL;
                        goto out;
                }
-               if (unlikely(!access_ok(vrfy_dir(type), buf, len))) {
+               if (check_access
+                   && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
                        ret = -EFAULT;
                        goto out;
                }
@@ -705,7 +707,7 @@ static ssize_t do_readv_writev(int type, struct file *file,
        }
 
        ret = rw_copy_check_uvector(type, uvector, nr_segs,
-                       ARRAY_SIZE(iovstack), iovstack, &iov);
+                                   ARRAY_SIZE(iovstack), iovstack, &iov, 1);
        if (ret <= 0)
                goto out;
 
index 8779405e15a871eba64681021065ed1ab0e71853..4ce46bd370cdaa8a76ad0ed6d984251c9b76de7d 100644 (file)
@@ -548,7 +548,8 @@ extern ssize_t compat_rw_copy_check_uvector(int type,
                const struct compat_iovec __user *uvector,
                unsigned long nr_segs,
                unsigned long fast_segs, struct iovec *fast_pointer,
-               struct iovec **ret_pointer);
+               struct iovec **ret_pointer,
+               int check_access);
 
 extern void __user *compat_alloc_user_space(unsigned long len);
 
index 178cdb4f1d4afe81a66e631872de4b587e95d1c4..e4fe6034b96a2f6cbf61cbd515d9d709faa62760 100644 (file)
@@ -1625,9 +1625,10 @@ struct inode_operations {
 struct seq_file;
 
 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
-                               unsigned long nr_segs, unsigned long fast_segs,
-                               struct iovec *fast_pointer,
-                               struct iovec **ret_pointer);
+                             unsigned long nr_segs, unsigned long fast_segs,
+                             struct iovec *fast_pointer,
+                             struct iovec **ret_pointer,
+                             int check_access);
 
 extern ssize_t vfs_read(struct file *, char __user *, size_t, loff_t *);
 extern ssize_t vfs_write(struct file *, const char __user *, size_t, loff_t *);
index 8c03b98df5f93d196d523c60f31a850ffe9a22e4..0dc0809125f63e11108420f562c6e32f1678e934 100644 (file)
@@ -847,4 +847,17 @@ asmlinkage long sys_open_by_handle_at(int mountdirfd,
                                      struct file_handle __user *handle,
                                      int flags);
 asmlinkage long sys_setns(int fd, int nstype);
+asmlinkage long sys_process_vm_readv(pid_t pid,
+                                    const struct iovec __user *lvec,
+                                    unsigned long liovcnt,
+                                    const struct iovec __user *rvec,
+                                    unsigned long riovcnt,
+                                    unsigned long flags);
+asmlinkage long sys_process_vm_writev(pid_t pid,
+                                     const struct iovec __user *lvec,
+                                     unsigned long liovcnt,
+                                     const struct iovec __user *rvec,
+                                     unsigned long riovcnt,
+                                     unsigned long flags);
+
 #endif
index 72c9e4fdde00d688648d54b153c614aed2f36f25..306742a28266e24469c89164be415e3d9fc52f4f 100644 (file)
@@ -5,7 +5,8 @@
 mmu-y                  := nommu.o
 mmu-$(CONFIG_MMU)      := fremap.o highmem.o madvise.o memory.o mincore.o \
                           mlock.o mmap.o mprotect.o mremap.o msync.o rmap.o \
-                          vmalloc.o pagewalk.o pgtable-generic.o
+                          vmalloc.o pagewalk.o pgtable-generic.o \
+                          process_vm_access.o
 
 obj-y                  := filemap.o mempool.o oom_kill.o fadvise.o \
                           maccess.o page_alloc.o page-writeback.o \
diff --git a/mm/process_vm_access.c b/mm/process_vm_access.c
new file mode 100644 (file)
index 0000000..0df7696
--- /dev/null
@@ -0,0 +1,446 @@
+/*
+ *  linux/mm/process_vm_access.c
+ *
+ *  Copyright (C) 2010-2011 Christopher Yeoh <cyeoh@au1.ibm.com>, IBM Corp.
+ */
+
+#include <linux/mm.h>
+#include <linux/uio.h>
+#include <linux/sched.h>
+#include <linux/highmem.h>
+#include <linux/ptrace.h>
+#include <linux/slab.h>
+#include <linux/syscalls.h>
+
+#ifdef CONFIG_COMPAT
+#include <linux/compat.h>
+#endif
+
+/*
+ * process_vm_rw_pages - read/write pages from task specified
+ * @task: task to read/write from
+ * @mm: mm for task
+ * @process_pages: struct pages area that can store at least
+ *  nr_pages_to_copy struct page pointers
+ * @pa: address of page in task to start copying from/to
+ * @start_offset: offset in page to start copying from/to
+ * @len: number of bytes to copy
+ * @lvec: iovec array specifying where to copy to/from
+ * @lvec_cnt: number of elements in iovec array
+ * @lvec_current: index in iovec array we are up to
+ * @lvec_offset: offset in bytes from current iovec iov_base we are up to
+ * @vm_write: 0 means copy from, 1 means copy to
+ * @nr_pages_to_copy: number of pages to copy
+ */
+static ssize_t process_vm_rw_pages(struct task_struct *task,
+                                  struct mm_struct *mm,
+                                  struct page **process_pages,
+                                  unsigned long pa,
+                                  unsigned long start_offset,
+                                  unsigned long len,
+                                  const struct iovec *lvec,
+                                  unsigned long lvec_cnt,
+                                  unsigned long *lvec_current,
+                                  size_t *lvec_offset,
+                                  int vm_write,
+                                  unsigned int nr_pages_to_copy)
+{
+       int pages_pinned;
+       void *target_kaddr;
+       int pgs_copied = 0;
+       int j;
+       int ret;
+       ssize_t bytes_to_copy;
+       ssize_t bytes_copied = 0;
+       ssize_t rc = -EFAULT;
+
+       /* Get the pages we're interested in */
+       down_read(&mm->mmap_sem);
+       pages_pinned = get_user_pages(task, mm, pa,
+                                     nr_pages_to_copy,
+                                     vm_write, 0, process_pages, NULL);
+       up_read(&mm->mmap_sem);
+
+       if (pages_pinned != nr_pages_to_copy)
+               goto end;
+
+       /* Do the copy for each page */
+       for (pgs_copied = 0;
+            (pgs_copied < nr_pages_to_copy) && (*lvec_current < lvec_cnt);
+            pgs_copied++) {
+               /* Make sure we have a non zero length iovec */
+               while (*lvec_current < lvec_cnt
+                      && lvec[*lvec_current].iov_len == 0)
+                       (*lvec_current)++;
+               if (*lvec_current == lvec_cnt)
+                       break;
+
+               /*
+                * Will copy smallest of:
+                * - bytes remaining in page
+                * - bytes remaining in destination iovec
+                */
+               bytes_to_copy = min_t(ssize_t, PAGE_SIZE - start_offset,
+                                     len - bytes_copied);
+               bytes_to_copy = min_t(ssize_t, bytes_to_copy,
+                                     lvec[*lvec_current].iov_len
+                                     - *lvec_offset);
+
+               target_kaddr = kmap(process_pages[pgs_copied]) + start_offset;
+
+               if (vm_write)
+                       ret = copy_from_user(target_kaddr,
+                                            lvec[*lvec_current].iov_base
+                                            + *lvec_offset,
+                                            bytes_to_copy);
+               else
+                       ret = copy_to_user(lvec[*lvec_current].iov_base
+                                          + *lvec_offset,
+                                          target_kaddr, bytes_to_copy);
+               kunmap(process_pages[pgs_copied]);
+               if (ret) {
+                       pgs_copied++;
+                       goto end;
+               }
+               bytes_copied += bytes_to_copy;
+               *lvec_offset += bytes_to_copy;
+               if (*lvec_offset == lvec[*lvec_current].iov_len) {
+                       /*
+                        * Need to copy remaining part of page into the
+                        * next iovec if there are any bytes left in page
+                        */
+                       (*lvec_current)++;
+                       *lvec_offset = 0;
+                       start_offset = (start_offset + bytes_to_copy)
+                               % PAGE_SIZE;
+                       if (start_offset)
+                               pgs_copied--;
+               } else {
+                       start_offset = 0;
+               }
+       }
+
+       rc = bytes_copied;
+
+end:
+       if (vm_write) {
+               for (j = 0; j < pages_pinned; j++) {
+                       if (j < pgs_copied)
+                               set_page_dirty_lock(process_pages[j]);
+                       put_page(process_pages[j]);
+               }
+       } else {
+               for (j = 0; j < pages_pinned; j++)
+                       put_page(process_pages[j]);
+       }
+
+       return rc;
+}
+
+/* Maximum number of pages kmalloc'd to hold struct page's during copy */
+#define PVM_MAX_KMALLOC_PAGES (PAGE_SIZE * 2)
+
+/*
+ * process_vm_rw_single_vec - read/write pages from task specified
+ * @addr: start memory address of target process
+ * @len: size of area to copy to/from
+ * @lvec: iovec array specifying where to copy to/from locally
+ * @lvec_cnt: number of elements in iovec array
+ * @lvec_current: index in iovec array we are up to
+ * @lvec_offset: offset in bytes from current iovec iov_base we are up to
+ * @process_pages: struct pages area that can store at least
+ *  nr_pages_to_copy struct page pointers
+ * @mm: mm for task
+ * @task: task to read/write from
+ * @vm_write: 0 means copy from, 1 means copy to
+ */
+static ssize_t process_vm_rw_single_vec(unsigned long addr,
+                                       unsigned long len,
+                                       const struct iovec *lvec,
+                                       unsigned long lvec_cnt,
+                                       unsigned long *lvec_current,
+                                       size_t *lvec_offset,
+                                       struct page **process_pages,
+                                       struct mm_struct *mm,
+                                       struct task_struct *task,
+                                       int vm_write)
+{
+       unsigned long pa = addr & PAGE_MASK;
+       unsigned long start_offset = addr - pa;
+       unsigned long nr_pages;
+       ssize_t bytes_copied = 0;
+       ssize_t rc;
+       unsigned long nr_pages_copied = 0;
+       unsigned long nr_pages_to_copy;
+       unsigned long max_pages_per_loop = PVM_MAX_KMALLOC_PAGES
+               / sizeof(struct pages *);
+
+
+       /* Work out address and page range required */
+       if (len == 0)
+               return 0;
+       nr_pages = (addr + len - 1) / PAGE_SIZE - addr / PAGE_SIZE + 1;
+
+
+       while ((nr_pages_copied < nr_pages) && (*lvec_current < lvec_cnt)) {
+               nr_pages_to_copy = min(nr_pages - nr_pages_copied,
+                                      max_pages_per_loop);
+
+               rc = process_vm_rw_pages(task, mm, process_pages, pa,
+                                        start_offset, len,
+                                        lvec, lvec_cnt,
+                                        lvec_current, lvec_offset,
+                                        vm_write, nr_pages_to_copy);
+               start_offset = 0;
+
+               if (rc < 0)
+                       return rc;
+               else {
+                       bytes_copied += rc;
+                       len -= rc;
+                       nr_pages_copied += nr_pages_to_copy;
+                       pa += nr_pages_to_copy * PAGE_SIZE;
+               }
+       }
+
+       rc = bytes_copied;
+       return rc;
+}
+
+static ssize_t process_vm_rw(pid_t pid, const struct iovec *lvec,
+                            unsigned long liovcnt,
+                            const struct iovec *rvec,
+                            unsigned long riovcnt,
+                            unsigned long flags, int vm_write)
+{
+       struct task_struct *task;
+       struct page **process_pages = NULL;
+       struct mm_struct *mm;
+       unsigned long i;
+       ssize_t rc;
+       ssize_t bytes_copied;
+       unsigned long nr_pages = 0;
+       unsigned long nr_pages_iov;
+       unsigned long iov_l_curr_idx = 0;
+       size_t iov_l_curr_offset = 0;
+       ssize_t iov_len;
+
+       /*
+        * Work out how many pages of struct pages we're going to need
+        * when eventually calling get_user_pages
+        */
+       for (i = 0; i < riovcnt; i++) {
+               iov_len = rvec[i].iov_len;
+               if (iov_len > 0) {
+                       nr_pages_iov = ((unsigned long)rvec[i].iov_base
+                                       + iov_len)
+                               / PAGE_SIZE - (unsigned long)rvec[i].iov_base
+                               / PAGE_SIZE + 1;
+                       nr_pages = max(nr_pages, nr_pages_iov);
+               }
+       }
+
+       if (nr_pages == 0)
+               return 0;
+
+       /* For reliability don't try to kmalloc more than 2 pages worth */
+       process_pages = kmalloc(min_t(size_t, PVM_MAX_KMALLOC_PAGES,
+                                     sizeof(struct pages *)*nr_pages),
+                               GFP_KERNEL);
+
+       if (!process_pages)
+               return -ENOMEM;
+
+       /* Get process information */
+       rcu_read_lock();
+       task = find_task_by_vpid(pid);
+       if (task)
+               get_task_struct(task);
+       rcu_read_unlock();
+       if (!task) {
+               rc = -ESRCH;
+               goto free_proc_pages;
+       }
+
+       task_lock(task);
+       if (__ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
+               task_unlock(task);
+               rc = -EPERM;
+               goto put_task_struct;
+       }
+       mm = task->mm;
+
+       if (!mm || (task->flags & PF_KTHREAD)) {
+               task_unlock(task);
+               rc = -EINVAL;
+               goto put_task_struct;
+       }
+
+       atomic_inc(&mm->mm_users);
+       task_unlock(task);
+
+       rc = 0;
+       for (i = 0; i < riovcnt && iov_l_curr_idx < liovcnt; i++) {
+               bytes_copied = process_vm_rw_single_vec(
+                       (unsigned long)rvec[i].iov_base, rvec[i].iov_len,
+                       lvec, liovcnt, &iov_l_curr_idx, &iov_l_curr_offset,
+                       process_pages, mm, task, vm_write);
+               if (bytes_copied < 0) {
+                       rc = bytes_copied;
+                       goto put_mm;
+               } else {
+                       rc += bytes_copied;
+               }
+       }
+
+put_mm:
+       mmput(mm);
+
+put_task_struct:
+       put_task_struct(task);
+
+
+free_proc_pages:
+       kfree(process_pages);
+       return rc;
+}
+
+static ssize_t process_vm_rw_check_iovecs(pid_t pid,
+                                         const struct iovec __user *lvec,
+                                         unsigned long liovcnt,
+                                         const struct iovec __user *rvec,
+                                         unsigned long riovcnt,
+                                         unsigned long flags, int vm_write)
+{
+       struct iovec iovstack_l[UIO_FASTIOV];
+       struct iovec iovstack_r[UIO_FASTIOV];
+       struct iovec *iov_l = iovstack_l;
+       struct iovec *iov_r = iovstack_r;
+       ssize_t rc;
+
+       if (flags != 0)
+               return -EINVAL;
+
+       /* Check iovecs */
+       if (vm_write)
+               rc = rw_copy_check_uvector(WRITE, lvec, liovcnt, UIO_FASTIOV,
+                                          iovstack_l, &iov_l, 1);
+       else
+               rc = rw_copy_check_uvector(READ, lvec, liovcnt, UIO_FASTIOV,
+                                          iovstack_l, &iov_l, 1);
+       if (rc <= 0)
+               goto free_iovecs;
+
+       rc = rw_copy_check_uvector(READ, rvec, riovcnt, UIO_FASTIOV,
+                                  iovstack_r, &iov_r, 0);
+       if (rc <= 0)
+               goto free_iovecs;
+
+       rc = process_vm_rw(pid, iov_l, liovcnt, iov_r, riovcnt, flags,
+                           vm_write);
+
+free_iovecs:
+       if (iov_r != iovstack_r)
+               kfree(iov_r);
+       if (iov_l != iovstack_l)
+               kfree(iov_l);
+
+       return rc;
+}
+
+SYSCALL_DEFINE6(process_vm_readv, pid_t, pid, const struct iovec __user *, lvec,
+               unsigned long, liovcnt, const struct iovec __user *, rvec,
+               unsigned long, riovcnt, unsigned long, flags)
+{
+       return process_vm_rw_check_iovecs(pid, lvec, liovcnt, rvec, riovcnt,
+                                         flags, 0);
+}
+
+SYSCALL_DEFINE6(process_vm_writev, pid_t, pid,
+               const struct iovec __user *, lvec,
+               unsigned long, liovcnt, const struct iovec __user *, rvec,
+               unsigned long, riovcnt, unsigned long, flags)
+{
+       return process_vm_rw_check_iovecs(pid, lvec, liovcnt, rvec, riovcnt,
+                                         flags, 1);
+}
+
+#ifdef CONFIG_COMPAT
+
+asmlinkage ssize_t
+compat_process_vm_rw_check_iovecs(compat_pid_t pid,
+                                 const struct compat_iovec __user *lvec,
+                                 unsigned long liovcnt,
+                                 const struct compat_iovec __user *rvec,
+                                 unsigned long riovcnt,
+                                 unsigned long flags, int vm_write)
+{
+       struct iovec iovstack_l[UIO_FASTIOV];
+       struct iovec iovstack_r[UIO_FASTIOV];
+       struct iovec *iov_l = iovstack_l;
+       struct iovec *iov_r = iovstack_r;
+       ssize_t rc = -EFAULT;
+
+       if (flags != 0)
+               return -EINVAL;
+
+       if (!access_ok(VERIFY_READ, lvec, liovcnt * sizeof(*lvec)))
+               goto out;
+
+       if (!access_ok(VERIFY_READ, rvec, riovcnt * sizeof(*rvec)))
+               goto out;
+
+       if (vm_write)
+               rc = compat_rw_copy_check_uvector(WRITE, lvec, liovcnt,
+                                                 UIO_FASTIOV, iovstack_l,
+                                                 &iov_l, 1);
+       else
+               rc = compat_rw_copy_check_uvector(READ, lvec, liovcnt,
+                                                 UIO_FASTIOV, iovstack_l,
+                                                 &iov_l, 1);
+       if (rc <= 0)
+               goto free_iovecs;
+       rc = compat_rw_copy_check_uvector(READ, rvec, riovcnt,
+                                         UIO_FASTIOV, iovstack_r,
+                                         &iov_r, 0);
+       if (rc <= 0)
+               goto free_iovecs;
+
+       rc = process_vm_rw(pid, iov_l, liovcnt, iov_r, riovcnt, flags,
+                           vm_write);
+
+free_iovecs:
+       if (iov_r != iovstack_r)
+               kfree(iov_r);
+       if (iov_l != iovstack_l)
+               kfree(iov_l);
+
+out:
+       return rc;
+}
+
+asmlinkage ssize_t
+compat_sys_process_vm_readv(compat_pid_t pid,
+                           const struct compat_iovec __user *lvec,
+                           unsigned long liovcnt,
+                           const struct compat_iovec __user *rvec,
+                           unsigned long riovcnt,
+                           unsigned long flags)
+{
+       return compat_process_vm_rw_check_iovecs(pid, lvec, liovcnt, rvec,
+                                                riovcnt, flags, 0);
+}
+
+asmlinkage ssize_t
+compat_sys_process_vm_writev(compat_pid_t pid,
+                            const struct compat_iovec __user *lvec,
+                            unsigned long liovcnt,
+                            const struct compat_iovec __user *rvec,
+                            unsigned long riovcnt,
+                            unsigned long flags)
+{
+       return compat_process_vm_rw_check_iovecs(pid, lvec, liovcnt, rvec,
+                                                riovcnt, flags, 1);
+}
+
+#endif
index 338b510e90275b1da873ed83aa907c404997d353..4c48e13448f84e0df200ab0a2f0895d0eb5a709c 100644 (file)
@@ -38,7 +38,7 @@ long compat_keyctl_instantiate_key_iov(
 
        ret = compat_rw_copy_check_uvector(WRITE, _payload_iov, ioc,
                                           ARRAY_SIZE(iovstack),
-                                          iovstack, &iov);
+                                          iovstack, &iov, 1);
        if (ret < 0)
                return ret;
        if (ret == 0)
index eca51918c951d3a28db16bf0f61c749722b6ac64..0b3f5d72af1cecbd06b33c2151ec570a4cfaa2ee 100644 (file)
@@ -1065,7 +1065,7 @@ long keyctl_instantiate_key_iov(key_serial_t id,
                goto no_payload;
 
        ret = rw_copy_check_uvector(WRITE, _payload_iov, ioc,
-                                   ARRAY_SIZE(iovstack), iovstack, &iov);
+                                   ARRAY_SIZE(iovstack), iovstack, &iov, 1);
        if (ret < 0)
                return ret;
        if (ret == 0)